Michelle, Thank you for forwarding Zadeh's fascinating note from the Soft Computing mailing list. My responses to several of his comments follow: > (1) Successes of probability theory mask a fundamental limitation -- the > inability to operate on what may be called perception-based information. I don't understand this comment. I can treat a belief about an event as a probability just as well as a frequency based estimate of the probability. To me, the distinction is between standard Bayes and empirical Bayes - primarily the minimum requirement to formulate a probability. The difficulty lies in extracting a probability that represents a belief from limited data. Further, I can easily extend an application and use a three factored probability logic (true-false-uncertain) or Dempster-Shafer reasoning, when the difficulty of gauging probabilities limits the solution to a problem. > (2) My examples are intended to challenge the unquestioned belief within > the Bayesian community that probability theory can handle any kind of > information, including information which is perception-based. In my limited experience, the community has lots of questions. It's more of the nature of an operating assumption. (If I haven't analyzed a problem successfully with a Bayesian approach, maybe I haven't tried hard enough.) > (3) The growing complexity of problems which arise in the conception, > design and utilization of information/intelligent systems requires that > all relevant methodologies be marshalled for solution. However, this statement might be true, even if Bayesian approaches were adequate. Perhaps all we need are Bayesian applications that are more clever or more sophisticated. Side-by-side demonstrations of how well each approach handles a problem would help me in these design formulation discussions. What does a new methodology require, can I use off-the-shelf software, what are the other resource requirements, what are the major assumptions, and what does an alternative yield, by way of improvement on a standard approach? > (4) What this points to is a need for formation of a coalition of > methodologies which, in combination, are much more powerful than they > are in a stand-alone mode. Yes, but I will want a clearer understanding of how to classify problems such that I know which techniques to apply first - which work best - a kind of operating manual. A smaller, but similar problem exists within applications of Bayesian probability theory: how do I chose the best probability distribution for the problem at hand? Right now, the available guidance is limited. > (5) Such a coalition or consortium is soft computing -- a synergistic > collection of computationally-oriented methodologies whose principal > members are fuzzy logic, neurocomputing, evolutionary computing, > probabilistic computing, chaotic computing and machine learning theory. > The obvious superiority of soft computing over any one of its > constituents suggests that, in the future, most information/intelligent > systems will be of hybrid type, employing various combination of > methodologies to deal with uncertainty, possibility, imprecision, > incompleteness, partial understanding and partial truth. A "coalition" may be closer than the note suggests. For example, even some confirmed Bayesians still take time to play with fuzzy logic or neural nets. However, Zadeh's suggested descriptor, "soft computing," does not seem appropriate. In practice, the codes for the algorithms of these procedures have just about the same degree of difficulty and that each yields the same output reproducably. Maybe he can invent a better term for the class. - --- Daniel M. Byrd III, Ph.D., D.A.B.T. President, Consultants in Toxicology, Risk Assessment and Product Safety Suite N707 560 N Street, SW Washington, DC 20024 (202)484-7707 - phone (202)484-0616 - fax [EMAIL PROTECTED] - email
